Resolving floral development dynamics using genome and single‐cell temporal transcriptome of Dendrobium devonianum

Summary

Dendrobium devonianum, a species of the Orchidaceae family, is notable for its unique floral characteristics, which include two yellow spots and purple tips on its labellum, as well as fringed edges. However, the molecular mechanisms underlying flower pattern formation in D. devonianum remain poorly understood, hindering advancements in its breeding process. Here, a chromosome‐scale genome of D. devonianum was presented for the first time, revealing two significant polyploidization events. Additionally, a high‐resolution single‐cell transcriptomic atlas was constructed, capturing 11 distinct cell clusters. Expression patterns of MADS‐box genes were identified through temporal and spatial bulk RNA‐Seq, revealing alignment with the ABCDE model of flower formation. Meanwhile, mass spectrometry imaging and scRNA analyses showed that the yellow spots were primarily associated with carotenoid biosynthesis gene expression, while the purple colour is predominantly linked to anthocyanin biosynthesis gene expression. These genes were mainly expressed in the epidermis and vascular cells. Developmental trajectory analyses of epidermal cells further uncovered a gene regulatory network and several transcription factors likely responsible for fringes formation along the labellum margin. This study provides valuable insights into the molecular mechanisms driving floral colour differentiation and structural traits in D. devonianum, contributing to a deeper understanding of orchid evolution, diversification and breeding.

Introduction

Orchidaceae, celebrated for its distinctive flower morphologies and ecological adaptations, is one of the most diverse plant families, comprising approximately 30,000 species (Li et al., 2021). Dendrobium, the second‐largest genus in Orchidaceae after Bulbophyllum, includes approximately 800–1500 species. Many Dendrobium species, such as D. candidum, D. nobile and D. devonianum, are valued as ornamental plants and traditional herbal medicines (Wei et al., 2024). Among them, D. devonianum has been a prized medicinal plant in China for over a century (Zhan et al., 2023). Its stem has traditionally been used as a tonic herb to treat hepatitis, asthma and immunological disorders (Deng et al., 2018).

Recognized as a national key secondary protected wild plant, D. devonianum is listed as an endangered species (EN) on the IUCN Red List of Threatened Species. This species exhibits distinctive floral characteristics, including a white labellum with a compound fringes margin and yellow spots on either side (Figure S1). The labellum's surface is densely populated with numerous conical and semicircular papillae, while the mesochile and epichile regions display jagged, branched epidermal cell structures and multicellular trichomes (Burzacka‐Hinz et al., 2022). The yellow coloration and fringes structure not only enhance the flower's visual appeal but also play a critical role in attracting and interacting with pollinators. Research indicates that several factors, including the composition, pigment content and spatiotemporal distribution of different pigments, influence flower coloration.

Dendrobium species exhibit a remarkable diversity in flower colours, morphologies and growth forms (Leng et al., 2024; Teixeira et al., 2014). Despite the vast diversity, Dendrobium orchids share a similar floral structure: three outer sepals, three inner petals, and a column formed by fused stamens and pistils (Zhao et al., 2023). The inner petals whorl usually consists of two lateral petals and a median labellum. Since all expected whorls in flowers are present in orchids, such a highly sophisticated flower organization offers an opportunity to discover new variant genes within morphogenetic networks. Recent studies on orchids have revealed that the MADS‐box family, responsible for forming the ABCDE model, plays a crucial role in orchid flowering and floral patterning (Li et al., 2022). Understanding the molecular mechanisms of flowering and floral development can be applied to diverse orchid breeding to achieve desired flowering traits and patterns.

Although advancements have been made in understanding orchid flower development, such as the genomic sequencing of related species like Cymbidium, Oncidium, Dendrobium and Phalaenopsis, the molecular mechanisms underlying the unique floral traits of D. devonianum remain poorly understood. Specifically, detailed insights into the formation of flower colour, fringed labellum structures and yellow‐spotted pigmentation are lacking due to the absence of comprehensive genomic and transcriptomic data. This gap in knowledge hinders our understanding of the genetic regulation of flower development in D. devonianum (Ai et al., 2023; Caì et al., 2015; Niu et al., 2021; Yan et al., 2015; Zhang et al., 2021).

This study aims to systematically investigate the molecular mechanisms of floral development and colour formation in D. devonianum. Using high‐quality genome assembly enabled by ultra‐long‐read sequencing and Hi‐C technology, combined with transcriptomic and mass spectrometry imaging techniques, the study explores the genetic basis of these traits. To elucidate the gene regulatory network involved in floral development and fringes formation, a single‐cell temporal transcriptome atlas was constructed for three different flowering stages of D. devonianum. These findings offer valuable insights into the mechanisms driving the fringes formation and labellum coloration, with significant implications for the breeding of orchids and Dendrobium cultivars.

Results and discussion

High‐quality genome assembly and genome evolution

A total of 67.07 Gb of Oxford Nanopore long reads and 42.49 Gb of Hi‐C sequencing reads (Table S1) were generated. Genome survey results estimated the size of the D. devonianum genome to be approximately 950.08 Mb, with a heterozygosity rate of 1.95% (Figure S2, Table S2). The assembled genome has a total length of 1.11 Gb, consisting of 326 contigs with a contig N50 of 9.03 Mb and the longest contig reaching 27.13 Mb (Figure 1a, Table S2). BUSCO analysis showed that the assembly completeness of the genome was 96.80% (Table S3). Furthermore, 96.74% of the contigs were anchored to 19 pseudochromosomes ranging in size from 36.02 to 82.88 Mb (Figure 1a, Figure S3, Table S4). The genome contains 829.68 Mb of repetitive sequences, accounting for 74.71% of the genome (Table S5). Among these, long terminal repeat retrotransposons were the most prevalent, with Copia elements (22.68%) being more abundant than Gypsy elements (15.79%) (Table S5). Additionally, 32,759 protein‐coding genes were predicted in the D. devonianum genome, with a BUSCO completeness estimation of 94.5% (Table S4).

Figure 1.

Genomic feature, syntenic analysis and phylogenetic positions of D. devonianum. (a) Genome characteristics of D. devonianum. The circos plot from the outer to the inner circle represents chromosome‐scale pseudochromosomes (Chr1–Chr19) (I), GC content (II), LTR number (III), Gypsy (IV), Copia (V), LAI (VI), gene density (VII), each linking line in the centre of the circos plot indicates a pair of homologous genes (VII), respectively. (b) Phylogenomic Analysis and Gene Family Distribution. The phylogenetic tree of plant species shows the expansion and contraction of gene families in D. devonianum and 15 other plant species. Divergence times for these species are indicated by numbers on the nodes (millions of years ago, Mya), and transparent blue bars at the internodes show the 95% highest posterior density (HPD). Pie charts depict the proportions of gene families that experienced contraction (red) or expansion (green). (c) Substitution‐rate‐adjusted mixed paralog‐ortholog synonymous substitutions per synonymous site (K _S) plot. This plot for D. devonianum, generated by ksrates, shows the anchor‐pair K _S distribution in grey, with two inferred WGD components indicated in blue, based on lognormal mixture model clustering. The vertical dashed lines labelled ‘a’ denote the modes of these components. Lines representing the same speciation event in the phylogeny share colour and numbering. The phylogram, produced by ksrates from the input phylogenetic tree, has branch lengths set according to the K _S distances estimated from ortholog K _S distributions. (d) Dot plots show the syntenic orthologs between D. devonianum and D. devonianum.

A phylogenomic tree was constructed using homologous genes from D. devonianum and 15 other plant species, with Amborella trichopoda serving as the outgroup (Table S6). To determine the phylogenetic position of D. devonianum and estimate its divergence times, a high‐confidence phylogenetic tree was constructed using 165 single‐copy gene families (Figure 1b). The analysis revealed that D. devonianum clustered with D. catenatum, D. chrysotoxum, D. huoshanense and D. nobile. The divergence time between D. devonianum and D. nobile was estimated at approximately 18.79 million years ago (Mya), with a 95% confidence interval (CI) of 15.27 to 23.77 Mya (Figure S4). Within the D. devonianum genome, 917 gene families were identified as expanded, while 965 gene families were contracted (Figure 1b). Functional annotation of the expanded gene families indicated enrichment in pathways related to the biosynthesis of secondary metabolites, plant organ formation and plant hormone signal transduction (Figure S5). Notably, expansions in flavonoid and carotenoid biosynthesis pathways were observed, encompassing genes encoding F3'5'H, OMT, ANR and BCH. Additionally, MADS‐box gene families exhibited expansions, particularly in the B sister clade genes. These findings highlight the role of gene family expansions in shaping the unique floral and metabolic traits of D. devonianum.

Polyploidization, or whole‐genome duplication (WGD), occurs widely in land plants, particularly ferns and angiosperms. This process leads to the emergence of novel and varied phenotypes. The K _S (the number of substitutions per synonymous site) distribution of intragenomic paralogs showed two clear peaks of duplicate genes at K _S values of approximately 0.89 (WGD1) and 1.5 (WGD2), indicating that D. devonianum likely underwent two polyploidization events (Figure 1c,d). The syntenic relationship shows that the collinearity blocks were mainly in a 1:1 pattern between D. devonianum and other Dendrobium genomes, suggesting no species‐specific WGD events had occurred in D. devonianum (Figure S6a–c). The K _S peaks of orthologous pairs between both D. devonianum/P. equestris and D. devonianum/D. chrysotoxum were below 0.89, suggesting that the WGD1 events occurred before the differentiation of these species (Figure 1c). Furthermore, the K _S peak (K _S ≈ 0.88) between D. devonianum and A. shenzhenica was slightly smaller than but close to the paralogous K _S peak of A. shenzhenica (K _S ≈ 1), indicating that extant orchid species share a single WGD event (Figure S7). These findings suggested that D. devonianum experienced two polyploidization events. The earlier WGD event was shared among most monocot species, and a later WGD event was shared among all extant orchid species, aligning with previous research (Niu et al., 2021; Xu et al., 2022). Further investigation revealed that both WGD and tandem duplication significantly contributed to the expansion of MADS‐box transcription factors, including members of the AP1, SEP, SOC1 and SVP clades, as well as pigment biosynthetic enzymes such as CHS, F3'H, LAR, UGT72L and CDD (Table S7). These duplications enhanced the copy number of key regulators governing floral organ identity and enzymes critical for flavonoid and carotenoid diversification.

Construction of single‐cell temporal transcriptome atlas of flowers

For the analysis of single‐cell RNA sequencing (scRNA‐seq) of D. devonianum flower cells, three flower samples (S1, S6 and S15) from different developmental stages were subjected to Illumina high‐throughput sequencing (Figure 2a). 377.19 Gb of scRNA‐seq data were produced (Table S8). Gene expression profiles derived from scRNA‐seq and bulk RNA‐seq datasets showed strong correlations (r > 0.8), suggesting the accurate recovery of the global transcriptome profile in D. devonianum flowers using scRNA‐seq data (Figure S12). To construct a cell atlas of flower development in D. devonianum, three flower samples were integrated for cell clustering and annotation. In total, 31 511 single cells were analysed to identify distinct cell populations (Table S9). Uniform manifold approximation and projection (UMAP) methods were applied to visualize the cells from the three time points, revealing 11 distinct clusters in two‐dimensional space (Figure 2b). These clusters differed across developmental stages regarding the proportion of cell types. Clusters 0, 1, 4 and 6 contained higher proportions of cells from the S1 and S6 samples, whereas clusters 2, 3 and 5 were dominated by cells from the S15 sample (Figure S9).

Figure 2.

Cluster annotation of single‐cell transcriptomes from flower cell populations. (a) The distribution of cells and phenotypes for flower samples at various developmental stages is visualized using UMAP. Orange dots represent cells from S1, blue dots represent cells from S6, and purple dots represent cells from S15. (b) The scRNA‐seq data of S1, S6 and S15 samples were combined for clustering. The UMAP graph visualizes 11 distinct cell clusters, with each dot representing an individual cell coloured according to its assigned cell type. Corresponding flower cluster IDs were indicated on the right. (c) The expression patterns of representative cell‐type marker genes across the cell clusters were displayed. The dot diameter represents the proportion of cells expressing a particular gene in each cluster, while the colour indicates the scaled average expression.

Through examining gene expression patterns across different clusters, highly expressed marker genes were identified for each cell cluster. Genes associated with epidermal function, such as DEFECTIVE IN CUTICULAR RIDGES (DCR), GLUTATHIONE S‐TRANSFERASE F SUBUNIT 1 (GSTF1), LONG‐CHAIN ACYL‐COA SYNTHASE 1/2 (LACS1/2), PROTODERMAL FACTOR 1 (PDF1), FIDDLEHEAD (FDH) and 3‐KETOACYL‐COA SYNTHASE 4 (KCS4), which were involved in cuticular wax and suberin biosynthesis, were highly enriched in cluster 1 and cluster 3 (Figure 2c and Figures S10 and S11). Cluster 1 predominantly comprised cells from the S1 stage, whereas cluster 3 was primarily consisted of cells from the S15 stage. These results suggest that epidermal cell (EC) regulatory genes exhibit stage‐specific differences during development. Vascular bundle cells (VB) were predominantly allocated to cluster 5, which expressed the auxin efflux transporter PINFORMED 1C (PIN1C), THREONINE ALDOLASE 2 (THA2) and EXORDIUM‐like 2 (EXL2). Previous studies have demonstrated that AtPIN1, BdPIN1 and THA2 play key roles in leaf vascular development (Fusi et al., 2024; Joshi et al., 2006). In contrast, cluster 2 showed low expression levels of both epidermis and vascular bundle marker genes, making it challenging to characterize the cell type in this cluster accurately. Subsequent sub‐clustering of the cluster 2 yielded five sub‐clusters, with sub‐cluster 0 (63.7%) and sub‐cluster 1 (33.1%) dominated (Figure S12a–c). Epidermal cell marker genes exhibited higher expression in sub‐cluster 0, while vascular cell marker genes were weakly expressed in both sub‐cluster 0 and sub‐cluster 1 (Figure S12d). The functional annotations for cluster 2 included ‘lipid metabolic process’, ‘fatty acid metabolic process’ and ‘acyl‐CoA metabolic process’, supporting the inference that cluster 2 primarily comprises epidermal cells (Figure S12e). Mesophyll cells (MC) were predominantly assigned to cluster 0, with high expression levels of light‐dependent and chloroplast‐related genes, including RUBISCO SMALL SUBUNIT 1 (RBCS1), LIGHT HARVESTING COMPLEX OF PHOTOSYSTEM I 4 (LHCA4), LIGHT HARVESTING COMPLEX OF PHOTOSYSTEM II 3 (LHCB3) and CHLOROPHYLL A/B BINDING PROTEIN (CAB40). These genes were enriched for the Gene Ontology (GO) terms ‘photosynthesis’ and ‘rhythmic process’ (Figure S11).

Additionally, master regulatory genes related to xylem development, including LIKE AUX 2 (LAX2), PHOSPHATR 1 (PHO1) and WALLS ARE THIN 1 (WAT1), were highly enriched in cluster 4, indicating this cluster represents a xylem cell (XY) population. Similarly, cluster 8 was identified as phloem (PH) cells due to the expression of phloem marker genes, such as WUSCHEL‐RELATED HOMEOBOX 8 (WOX8), HEAT SHOCK PROTEIN 18 (HSP18), and PHLOEM PROTEIN 2‐LIKE A15 (PP2A15). Ovule cells (OV) in cluster 9 were marked by high expression of SPATULA (SPT) and AGAMOUS 2 (AG2). SPT is essential for the growth and development of carpel margin tissues, including the style, stigma, septum and transmitting tract (Makkena and Lamb, 2013). Notably, cluster 10 was designated as anther cells (AN) due to the specific expression of genes critical for pollen wall formation, including CYP704B1, CYP703A3, POLYKETIDE SYNTHASE A (PKSA) and TETRAKETIDE α‐PYRONE REDUCTASE (TKPR). These genes were enriched for GO terms related to pollen development and anther wall tapetum development. Guard cell (GC) marker genes, such as MPK10/14 and OPEN STOMATA 1A (OST1A), were highly expressed in cluster 7. Additionally, cluster 6 exhibited significant enrichment for cell cycle‐related genes, such as AUR1, CYCLIN‐DEPENDENT KINASE B2‐1 (CDKB2‐1), CYCA2‐1 and CYCLIN B2‐4 (CYCB2‐4). This cluster, designated as proliferating cells (PC), was enriched for GO terms including ‘chromosome segregation’, ‘cell division’ and ‘cell cycle process’ (Figure S11). In summary, our single‐cell transcriptome dataset identified 11 distinct clusters corresponding to nine major cell types, providing valuable insights into cell‐specific gene expression patterns and functions.

Classification and temporal–spatial expression of MADS‐box genes

The Orchidaceae family features a unique floral organ, the labellum (or lip), located in the second whorl. Previous studies have shown that floral organ identity is regulated by MADS‐box genes, which play a key role in the well‐characterized ABCDE model of flower development (Li et al., 2022; Lu et al., 2019). These genes were essential for numerous developmental processes in plants, particularly those related to flowers (Wang et al., 2019). Herein, 59 MADS‐box genes were identified in the D. devonianum genome (Table S10), divided into two clades: type I and type II. Among them, seven genes belong to the Mα and Mγ subgroups of type I, similar to other orchid species. Notably, none of the orchid species examined in this study possesses the Mβ subgroup (Figure S13), consistent with previous reports (Lu et al., 2019).

Moreover, the remaining genes were classified into the MIKCC and MIKC* subclades within the type II clades. The MIKCC subclades were further divided into 12 subfamilies: AGL2, AGL6, AP1/FUL, AG, STK, SOC1, ANR1/AGL17, Bs, AP3/PI, MIKC*, MADS32‐like and SVP (Figure 3a, Figure S14). The result revealed that most genes in the type II MADS‐box clades have undergone duplication, except for those in the AP1, AG, SVP subfamilies and the MIKC* subclades. Furthermore, we analysed the MADS‐box genes across all available orchid genomes. Gene duplication appears to be widespread among type II MADS genes in orchids. Most ABCDE clades within MADS‐box genes exhibit conserved copy numbers among closely related species; however, some species display significant expansions in specific clades (e.g. AP1/FUL in D catenatum, SEP/ANR clade in P. equestris and AG‐like in Gastrodia elata) (Table S11). Notably, the FLC, AGL12 and AGL15 clade were not identified in D. devonianum. Previous studies have also reported their absence in several orchid species, such as D. catenatum, D. chrysotoxum, P. equestris and C. ensifolium (Ai et al., 2023; Caì et al., 2015; Zhang et al., 2016, 2021). Together, the expansions and losses of MADS‐box genes highlight their role in shaping orchid floral evolution and adaptation.

Figure 3.

Phylogenetic analysis of MADS‐box genes and MADS‐box model in D. devonianum. (a) The phylogenetic tree of type I and type II MADS‐box genes in D. devonianum. The MADS‐box genes in D. devonianum are marked by a red solid cycle. (b) Expression patterns of MADS‐box genes across different development stages of D. devonianum flower (S1‐S15). (c) Expression patterns of MADS‐box genes in individual floral organs. These organs include the sepal (se), petal (pe), labellum (la), and column (co). (d) MADS‐box model of D. devonianum flowers. AP3: DedeAP3‐1, DedeAP3‐2, DedeAP3‐3, and DedeAP3‐4. PI: DedePI1. SEP: DedeSEP1, DedeSEP2, DedeSEP3 and DedeSEP4. AP1: DedeAP1‐1, DedeAP1‐5 and DedeAP1‐6. AP2: DedeAP2‐2, DedeAP2‐6 and DedeAP2‐13. AGL6‐1: DedeAGL6‐1. AGL6‐2: DedeAGL6‐2. AG: DedeAG1‐1, DedeAG1‐2, DedeAG2‐1 and DedeAG3. SOC: DedeSOC1‐1. STK: DedeSTK. The rectangles in different colours (pink, green, red, orange and purple) indicate that the respective genes were expressed in the floral organs, while a white rectangle signifies either low expression or non‐expression in those organs.

The ABC model proposes that a specific combination of MADS domain transcription factors (TFs) governs floral organ identity. This model was later expanded to include D and E homeotic genes, forming the ABCDE model (Zhao et al., 2023). Analysis of MADS‐box gene expression revealed that these genes can be classified into three distinct stages: inflorescence buds (S1–S5), pre‐anthesis (S6–S9) and anthesis (S10–S15) (Figure 3b). During the inflorescence bud stage, flowering‐related genes such as DedeAP1‐1, DedeAGL6‐1 and DedeSVP1 exhibited high expression levels. At the pre‐anthesis stage, DedeAGL6‐2, DedeAG1‐2, DedeAP3‐3, DedePI1 and DedeSVP2 were up‐regulated. In contrast, genes from the AP2‐like, SEP‐like and AP1‐like were significantly up‐regulated during the anthesis stage. These temporal shifts highlight stage‐specific regulatory roles for MADS‐box genes in floral development.

Spatial expression analysis further revealed that DedeAP1‐6 (class A), DedeAP3‐1, DedePI1 (class B) and DedeSEP3 (class E) were expressed across all three floral whorls. In contrast, DedeAG1 and DedeAG3 (class C), and DedeSTK (class D) were exclusively expressed in the column, consistent with their roles in reproductive organ development (Figure 3c, Figure S15). DcOAP2, a class A gene, was expressed in all floral organs (Yu and Goh, 2000), while DcOAG2, a homolog of A. thaliana SEEDSTICK (STK), was expressed primarily in ovaries and anthers (Xu et al., 2010). Notably, the AP3/DEF paralogs (DedeAP3‐1/‐2 and DedeAP3‐3/‐4), which underwent two duplication events (Figure S16), showed divergent spatial expression: the former in sepals, petals, and labellum, and the latter predominantly in mature labellum. Additionally, DedeAGL6‐1 and DedeAGL6‐2 exhibited organ‐specific expression in sepals/petals and labellum, respectively (Figure 3c, Figure S15). These patterns suggest a broad functional role for these genes in flower development. These findings support the established framework of floral organ identity in orchids (Li et al., 2022; Zhang et al., 2023).

In addition, single‐cell transcriptomic data further elucidated the spatial and temporal dynamics of the MADS‐box gene (Figure S17a,b). During late developmental stages (S15), genes such as DedeAP2‐4, DedeAP3‐2 and DedeAGL6‐2 were predominantly expressed in EC and VB cells. While DedeAGL65 and DedeAGL80 were highly active in PC during the early stages (S1 and S6), implicating them in cell division processes. Meanwhile, DedeAG1‐1, DedeAG2‐1 and DedeSVP1 showed elevated expression in OV cells across stages S1 and S15, consistent with their involvement in reproductive tissue development (Figure S17). These findings suggest that MADS‐box genes play specialized roles in regulating various aspects of flower formation, including floral organ identity, cell division and ovary development.

Overall, MADS‐box genes exhibit coordinated temporal and spatial expression patterns during floral development. Class A/B/E genes regulate perianth identity across all floral whorls, while class C/D genes specialize in column‐specific reproductive functions. The AGL6 genes expressed a similar pattern to those of AP1, AP3/PI and SEP, which may also have a potential role in petal formation. Single‐cell analysis further delineated their roles in cell‐type‐specific processes, from early proliferation to late maturation and ovary development, providing valuable insights into the spatiotemporal dynamics of MADS‐box gene regulation. These results not only support the established model of flower formation in orchids but also lay the groundwork for further molecular studies.

Labellum blotch coloration during flower development

Dendrobium devonianum flowers are distinguished by their labellum, which is white with a purple front, purplish red stripes on both sides below the middle and a yellow spot on either side (Figure 4a). Herein, the anthocyanins and carotenoid compounds present in the labellum of D. devonianum were determined by MALDI‐MSI. The results revealed that the concentrations of anthocyanins, including cyanidin‐3‐rutinoside, cyanidin‐3‐O‐(6‐O‐malonyl)‐glucoside, petunidin‐3‐O‐(6‐O‐acetyl)‐5‐O‐diglucoside and kaempferol‐glucoside, were significantly higher in the purple region compared to the yellow and white regions. Conversely, carotenoids, such as zeaxanthin and violaxanthin, were more concentrated in the yellow region compared to the purple and white regions (Figure 4c, Figure S18).

Figure 4.

Expression patterns of genes involved in the biosynthesis pathway of flavonoids and carotenoids in D. devonianum. (a) A schematic diagram of the labellum in D. devonianum. (b) Expression profiles of structural genes involved in the biosynthesis pathways of flavonoids and carotenoids are shown across different areas of the labellum. ABP, anthocyanin biosynthesis pathway; CBP, carotenoid biosynthesis pathway; la‐F, Labellum‐Fringes; la‐P, Labellum‐Purple; la‐W, Labellum‐White; la‐Y, Labellum‐Yellow. (c) MALDI‐MSI images heatmaps depict the accumulation patterns of metabolites within the labellum. (d) The expression level of putative genes encoding key enzymes involved in flavonoid and carotenoid biosynthesis across various developmental stages of floral development (S1–S15). The colour scale reflects the relative abundance of DEGs. (e) The heatmap displayed flavonoid and carotenoid biosynthetic genes specifically expressed in different cell types.

Moreover, the high‐quality genome was further utilized to identify key enzyme genes involved in the flower pigmentation pathway. The analysis identified 45 genes associated with the carotenoid biosynthesis pathway (CBP) and 71 genes involved in the anthocyanin/flavonoid biosynthesis pathway (ABP) (Table S12, Figure S19). Spatial expression analysis revealed distinct localization patterns. The carotenoid biosynthesis genes, including DedeCYP97C1, DedeBCH2, DedeCRTISO1, DedeCYP97B3, DedeLCYE1, DedePDS1 and DedePSY2, were significantly up‐regulated in the yellow blotches of the labellum (Figure 4b, Figure S20), consistent with elevated levels of zeaxanthin and violaxanthin. Previous studies have shown that PSY, PDS, ZEP and BCH were significantly up‐regulated in the yellow regions of plants (Chiou et al., 2010; Li et al., 2022; Sun et al., 2021). Conversely, genes associated with the ABP, such as DedeOMT2, DedeUGT72L7 and DedeLDOX3, were significantly up‐regulated in the purple regions of the labellum, promoting the formation of the purple colour (Figure 4b, Figure S20).

Temporal profiling of ABP and CBP genes across developmental stages (S1–S15) highlighted dynamic expression patterns. Notably, DedeOMT4, DedeCHI3, DedeF3H1, Dede4CL3, DedeCHS3, DedeF3GT, DedeFLS2, DedePAL2 and DedeLAR1 were highly expressed during the middle stages of development. In contrast, several genes, including DedeF3H3, DedeOMT9, DedeA5GT2, DedeUGT72L5, DedeZDS2, DedeCDD1, DedeCYP97A and DedeCRTISO1, were more highly expressed in the later stages (Figure 4d). These trends suggest coordinated activation of pigment biosynthesis during flower maturation. Additionally, single‐cell transcriptomics further localized key genes to specific cell types: DedeF3'H1, DedeFLS2, DedeBCH1, DedeFNS1 and DedeCYP97A were predominantly expressed in EC, whereas DedeLCYE1, DedeCRTISO1 and DedeFNS3 were enriched in VB at stage S15 (Figure 4e, Figure S21). KEGG enrichment confirmed flavonoid and carotenoid pathways in these cell clusters (Figure S22), supporting their roles in spatially restricted pigment synthesis.

The TFs serve as primary regulators of anthocyanin and carotenoid biosynthesis genes (Wang et al., 2022). A total of 1218 TFs were identified in D. devonianum (Table S13). The result suggested that DedeZDS2, DedeBCH1, DedeCDD2, DedePDS1 and Dede4CL1 may be regulated by transcription factors such as MYBs, AP2/ERFs, bHLHs, NACs, bZIPs and WRKYs (Figure S23a). MYB and bHLH families emerged as critical regulators (Table S14). Among them, DedeMYB305‐1, DedeMYB308, DedebHLH62‐2, DedebHLH104 and DedebHLH94‐1 showed strong positive correlations (|r| > 0.9, P < 0.05) with one or more structural genes and were highly expressed during late developmental stages. In contrast, DedeMYB17, DedebHLH93 and DedebHLH094‐2 exhibited negative correlations with structural genes, potentially acting as repressors (Figure S23b,c). Previous studies have shown that DchMYB2 and DchbHLH1 positively correlate with DchF3′H, DchF3′5′H, DchDFR and DchANS in Dendrobium, while OgMYB activates OgCHI and OgDFR in Oncidium (Hou et al., 2023; Li et al., 2017; Wang et al., 2022). These TFs may directly or indirectly regulate anthocyanin and carotenoid biosynthesis by modulating structural genes in the D. devonianum flower.

Overall, the floral pigmentation of D. devonianum arises from spatiotemporally regulated biosynthetic pathways. Carotenoid biosynthesis genes were predominant in the yellow labellum regions, with epidermal and vascular cell activity potentially facilitating zeaxanthin and violaxanthin accumulation. Anthocyanin biosynthesis genes drive purple coloration by promoting anthocyanin accumulation in the labellum. Co‐expression network analysis further suggested that MYB/bHLH TFs (e.g. DedeMYB305‐1 and DedebHLH104) may be key regulators, coordinating pigment‐related gene activity across developmental stages. This integrated framework of spatial expression, temporal activation dynamics and transcriptional regulation advances our understanding of the molecular mechanisms underlying floral coloration in orchids.

The developmental trajectory of fringes

The formation of fringes in D. devonianum likely involves the differentiation of multicellular trichomes at the labellum margin, though the cellular origins of this process remain uncharacterized (Burzacka‐Hinz et al., 2022). Trichomes are derived from epidermal cells. scRNA‐seq captures cells in both terminal and intermediate developmental states, allowing for exploring continuous differentiation trajectories within a developmental process. Re‐clustering the epidermal cell population revealed five sub‐clusters (EC_0, EC_1, EC_2, EC_3 and EC_4) (Figure 5a). Epidermis marker genes, including FDH, KCS and LACS, were highly expressed in the EC_1 and EC_4 clusters, suggesting these may represent epidermal cells. Notably, EC_2 and EC_3 predominantly consisted of trichome cells (TR), characterized by high expression levels of key genes such as GLABRA1 (GL1) and GLABRA3 (GL3) (Zhao et al., 2024). In contrast, EC_0 contained guard cells (GC), which exhibited high expression of the MYB60 protein, the bHLH transcription factor FAMA and SCREAM (SCRM) (Figure 5b).

Figure 5.

Developmental trajectory of epidermis cells. (a) UMAP visualization and cluster distribution of epidermis cell (EC) sub‐clusters. Each dot represents a single cell, with colours indicating distinct cell identities. (b) A dot plot illustrates the expression patterns of marker genes for EC, trichome (TR) and guard cell (GC) sub‐clusters. (c, d) The pseudotime trajectory depicts the sub‐clusters of EC, showing cell ordering along the differentiation trajectory represented by pseudotime states and cell types. (e) Expression of typical genes in epidermis fate cells along pseudotime. (f) Cell‐specific expression profiles of genes in the UMAP projection. (g) Trends in gene expression are analysed based on different differentiation fates. A heatmap displays the expression of the top 100 DEGs in the developmental trajectories of trichome and guard cells. (h) Heatmap shows the expression of the differentially expressed TFs in the developmental trajectories of trichome and guard cell sub‐clusters. The branch‐dependent genes exhibited three distinct differentiation patterns. Representative genes are listed on the right. The colour bar indicates the relative expression levels.

Pseudotime analysis was performed on the sub‐clusters EC_0 to EC_4, revealing a clear developmental trajectory that begins with cells from the EC subset and gradually diverges into two differentiated cell types: trichome cells and differentiated guard cells (Figure 5c,d). GL1‐1, MADS6, KAS12 and ABCG36 exhibit similar pseudotime curves (Figure 5e). Additionally, MYB305, NAC068 and BZIP12, etc., were highly expressed in TF clusters 5 (C5) (Figure 5h). GL1/3, KAS2/12 and ABCG36 were highly expressed in gene cluster 1 (C1) and cluster 4 (C4) (Figure 5e,g). Moreover, GL1‐1, GL3, KAS12 and MADS6 exhibited highly expressed in cell clusters at the later stages of flower development (S15) (Figure 5f). These findings suggest that these genes or TFs may play a significant role in fringes development. Moreover, these genes were enriched for the GO terms related to ‘lipid biosynthetic process’, ‘multicellular organism development’ and ‘plant organ development’ (Figure S24).

This pseudotime analysis provides novel insights into the dynamic processes underlying fringes development. GL1 and GL3 play a crucial role in regulating trichome differentiation in plants (Zhao et al., 2024). Similarly, KAS12, KAS2 and MADS6, which share comparable expression trajectories, may contribute to the structural integrity and functionality of fringes. Additionally, both GL1 and MADS6 are also expressed in vascular‐related clusters (Figure 5f). The potential cellular origins of the fringes may originate from the development and extension of epidermal or vascular cells. The precise mechanisms underlying fringes formation remain to be explored through further research. Establishing a transgenic system involved in fringes development will ultimately lead to a better understanding of these unique fringes.

Conclusions

D. devonianum was extensively utilized as an important medicinal plant and a healthy food source. This study constructed a high‐quality genome and scRNA‐Seq atlas of flowers at three developmental stages in D. devonianum, and nine distinct cell types were identified. Temporal and spatial bulk RNA‐Seq analyses were employed to identify differentially expressed MADS‐box genes, as well as carotenoid and anthocyanin biosynthesis genes, revealing evolutionary adaptations related to the floral organ identities in D. devonianum. Furthermore, carotenoid and anthocyanin biosynthesis genes displayed distinct expression patterns in the yellow spots and purple regions of the labellum. Pseudotime trajectory analysis of epidermal cells provided insights that could aid in discovering key regulators of epidermal cell differentiation. The multi‐omics dataset of D. devonianum represents a valuable resource for gene discovery and elucidating the molecular mechanisms underlying flower development.

Experimental procedures

Genome sequencing, survey and assembly

Fresh young leaves of D. devonianum were collected from Yunnan Province, China, for whole‐genome sequencing. Genomic DNA was extracted using the DNeasy Plant Mini Kit following the manufacturer's instructions. The quantity of DNA was measured with a NanoDrop 2000 spectrophotometer and a Qubit dsDNA HS assay kit (Thermo Fisher Scientific, USA). DNA fragments larger than 2 kb were purified using the BluePippin automatic nucleic acid electrophoresis and fragment recovery system (Sage Science, USA). These purified fragments were subsequently utilized to construct libraries for whole‐genome sequencing on the Nanopore PromethION platform (Oxford Nanopore Technologies, UK) at Biomarker Technologies, Beijing, China. Additionally, cells were extracted according to Phase Genomics protocols and sent to Phase Genomics (Seattle, WA, USA) for Hi‐C library preparation. The Hi‐C library was generated using the DPNII restriction enzyme and sequenced with a 150 bp read length on an Illumina HiSeq platform.

For transcriptome sequencing, total RNA was extracted from root, stem, leaf, flowers at various developmental stages (spanning from inflorescence buds to the blooming stage), different flower organs (column, labellum, petal, sepal) and different areas of the labellum (fringes, yellow, purple, white) using the RNAprep Pure Plant Kit (TIANGEN) (Table S15). The quality of the RNA was assessed on 1% agarose gels, while concentration and integrity were evaluated using a Qubit RNA Assay Kit and an Agilent 2100 Bioanalyzer (Agilent Technologies). Qualified RNA samples (3 μg each) were utilized to construct Illumina sequencing libraries. RNA sequencing was performed on the Illumina HiSeq 2500 platform at Novogene, generating 150 bp paired‐end reads following the manufacturer's protocol. Raw reads were processed using fastp to remove low‐quality sequences, and clean reads were aligned to the reference genome with HISAT2 (v2.2.1) using default parameters, retaining only uniquely mapped reads (Pertea et al., 2016). Gene expression levels were estimated with StringTie (v2.1.5) as FPKM (Fragments per kilobase of exon model per million reads mapped). Genes with FPKM >0.5 were considered as expressed and included in subsequent analyses. Differentially expressed genes (DEGs) were identified using DESeq2 (v1.28.1) with default parameters, applying a P‐value threshold of ≤0.01 and a fold change cut‐off of ≥2 (Love et al., 2014).

K‐mer frequency analysis was performed using Jellyfish (v2.0) to estimate genome size, heterozygosity and the proportion of repeat sequences. The D. devonianum genome was assembled with ONT long reads using NextDenovo (v2.5.2) and subsequently polished with Illumina DNA short reads via NextPolish (v1.3.1) to enhance base accuracy, with default parameters (Hu et al., 2021, 2024). Contig locations and orientations were initially determined by 3d‐DNA (v.180922) with default parameters. The resulting Hi‐C contact matrix was visualized using Juicebox (v1.11.08), where manual corrections were applied to rectify contig misassemblies and scaffold misjoins based on neighbouring interactions (Durand et al., 2016). Chromosomes were renamed and ordered by size, with homoeologous chromosomes numbered consecutively. Finally, the completeness and quality of the assembled genome were assessed using BUSCO (v5.4.7), leveraging gene content from the Embryophyta_odb10 database (Manni et al., 2021).

Gene annotations

De novo repeat identification was performed using RepeatModeler (v1.0.10), which employed two complementary computational methods, RECON v1.08 and RepeatScout v1.0.5, to delineate repeat element boundaries and infer family relationships from sequence data (Flynn et al., 2020). The resulting consensus repeat sequences merged and used to characterize transposable elements (TEs) through RepeatMasker (v4.0.7). Ab initio prediction was executed using the BRAKER (v2.1.5) pipeline, which automatically trained the gene predictors Augustus (v3.3.4) and GeneMark‐ET (Bruna et al., 2021). The pipeline integrated mapped transcriptome data with protein homology information to enhance prediction accuracy. For transcriptome‐based annotation, RNA‐seq data were aligned to the genome assembly using HISAT2, generating transcriptome alignments in BAM format. The final quality of the gene set was assessed using BUSCO to assess its completeness and accuracy.

Comparative genomic analysis

Orthologous groups, including D. devonianum and fourteen other sequenced species (Allium sativum, Apostasia shenzhenica, Asparagus officinalis, D. catenatum, D. chrysotoxum, D. huoshanense, D. nobile, Nymphaea colorata, Oryza sativa, Phalaenopsis equestris, Phoenix dactylifera, Solanum lycopersicum, Vitis vinifera and Zea mays), along with one outgroup species (Amborella trichopoda), were utilized for phylogenetic analysis. Single‐copy families across these species were identified using OrthoFinder (v2.3.11) with default parameters (Emms and Kelly, 2019). A phylogenetic tree comprising 221 single‐copy orthologous genes was constructed employing the maximum‐likelihood method in RAxML (v8.0.17) with 1000 bootstrap replicates (Stamatakis, 2006). Divergence times were estimated using MCMCTREE in PAML (v4.9) with approximate likelihood calculation. Calibration points for divergence events, including D. catenatum–P. equestris (51.0–123.0 Mya), A. shenzhenica–A. officinalis (92.5–118.5 Mya), D. catenatum–P. dactylifera (108.4–123.6 Mya), D. catenatum–V. vinifera (141.2–153.5 Mya) and D. catenatum–A. trichopoda (179.9–205.0 Mya), were obtained from the TimeTree database (Yang, 2007).

Gene family expansions and contractions were inferred using CAFE (v5.1.0), with a family‐wide P‐value <0.05 and a Viterbi P‐value <0.05 deemed indicative of significant expansion or contraction (Bie et al., 2006). Visualization was performed using Python scripts and iTOL. An enrichment analysis based on the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases was conducted to elucidate the biological functions of the expanded gene families. Functional annotations for the expanded and contracted genes were performed using the R package ClusterProfiler (v4.0) (Yu et al., 2012).

Whole‐genome duplication analysis

Syntenic block analysis, both within and between genomes, was conducted using MCScanX v1.1 with default parameters (Wang et al., 2012). The JCVI packages were employed to visualize syntenic relationships. The distribution analysis of Ks (substitutions per synonymous site) identified whole‐genome duplication (WGD) events. DIAMOND facilitated the self‐alignment of protein sequences from P. equestris, P. aphrodite, D. catenatum, A. shenzhenica, C. ensifolium and A. officinalis, as well as the extraction of optimal alignments. The ksrates tool (https://github.com/VIB‐PSB/ksrates) was used to identify WGD events by generating adjusted, mixed plots of paralog and ortholog Ks distributions (Sensalari et al., 2022). These plots aided in assessing the relative phylogenetic positioning of WGD and speciation events. Genome‐wide duplications in D. devonianum were classified into five categories using DupGen_Finder (v1.12) with default parameters (https://github.com/qiao‐xin/DupGen_finder).

MADS‐box gene family analysis

MADS‐box transcription factors in D. devonianum were identified using the MADS‐box protein sequences from A. thaliana and the HMMER (v3.0) profile (PF00319) (Finn et al., 2011). HMMER searches were conducted against all predicted proteins of D. devonianum with an E‐value threshold of 1e−10. The presence of the MADS domain was verified using the NCBI Conserved Domain Database, which can be accessed at http://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi (Aron et al., 2011). Additionally, the Simple Modular Architecture Research Tool was employed to validate the protein sequences. Subsequently, the MADS‐box protein sequences of C. ensifolium, P. equestris, A. thaliana and O. sativa were aligned using MAFFT (v7.525), and a phylogenetic tree was constructed using IQTREE (v2.3.6) (Lam‐Tung et al., 2015). Phylogenetic analyses incorporated ABCDE model proteins from orchid plants.

Identification of genes and TFs involved in biosynthesis

In the investigation of candidate genes involved in anthocyanin and flavonoid biosynthesis within the D. devonianum genome, a homology search was conducted using BLAST (v2.11.0) (Scott and Madden, 2004). Protein sequences from A. thaliana served as query sequences with an E‐value cut‐off of 1e‐10. A gene expression and co‐expression network was constructed to explore functional relationships by integrating transcriptome data with the identified biosynthesis genes and TFs. The TFs were identified using PlantTFDB with default parameters, and the resulting co‐expression networks were visualized using Cytoscape (v3.7.1) (Jin et al., 2014; Shannon et al., 2003).

scRNA‐seq library construction and sequencing

To study transcriptional dynamics during floral development, we performed scRNA‐seq on D. devonianum flowers at three developmental stages. The developmental stages S1 (early buds, ~0.2 cm, pre‐fringes formation), S6 (middle buds, fringes initiation, ~0.9 cm) and S15 (mature fully open flowers) were selected based on transcriptome clustering and critical developmental transitions (Figure 3a). Following the manufacturer's instructions, scRNA‐seq libraries were prepared using the 10× Genomics GemCode Single‐cell Platform. Protoplasts were introduced onto a Chromium microfluidic chip to generate single‐cell gel beads in emulsion (GEMs). The scRNA‐seq libraries were subsequently prepared using the Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 (10× Genomics). Library quality was assessed using an Agilent 2100 Bioanalyzer, and sequencing was performed on the Illumina NovaSeq 6000 platform.

Cell clustering and scRNA‐seq data analysis

The alignment of raw scRNA‐seq data was conducted against the D. devonianum genome. Raw BCL files were converted into FASTQ files, followed by alignment and quantification using 10x Genomics Cell Ranger (v3.1.0). Cell‐by‐gene matrices for each sample were subsequently imported into Seurat (v3.1.1) for further analysis. Initial quality control involved the removal of cells exhibiting unusually high unique molecular identifier (UMI) counts (≥8000), fewer than 500 or more than 4000 detected genes, or a mitochondrial gene percentage (≥10%). Doublets defined as multiple cells encapsulated within a single oil droplet were detected and excluded using DoubletFinder (v4.1.2). After filtering, the remaining cells underwent dimensionality reduction and clustering analysis. The resulting matrix was globally scaled and normalized using the LogNormalize method. Previously identified marker genes for distinct cell types were employed to classify three groups of high‐quality cells, facilitating the classification of cell clusters (Kang et al., 2022; Sun et al., 2023). GO and KEGG analyses were conducted to elucidate potential regulators within these clusters.

For cell clustering, normalized and scaled data were used to identify highly variable features. The top 2000 genes were selected and subjected to PCA for linear dimensionality reduction. Cell clusters were identified using the Seurat function FindClusters with a resolution of 0.3, utilizing the first 20 principal components (PCs). Subsequently, Uniform manifold approximation and projection (UMAP) visualization was applied to explore and present these clusters (Becht et al., 2019). UMAP was performed with 15 nearest neighbours, a minimum distance of 0.5, and an Euclidean distance metric. Clusters were defined using the Louvain algorithm with a resolution parameter of 0.1. Clusters were annotated by examining the expression of known marker genes. The FindMarkers function was employed to identify genes uniquely expressed in each cell cluster or population (Ryu et al., 2019). Candidate genes were further refined based on specific thresholds, including avg_log2 fold change (>0.5), adjusted P‐value (<0.01) and pct.1/pct.2 (>2). Monocle2 (v2.34.0) and Slingshot (v2.14.0) were employed to infer a continuous developmental trajectory (Kelly et al., 2018; Reid and Wernisch, 2016). Cluster‐enriched genes were utilized to select and order cells of interest, and the data were split into two components using the DDRTree approach. Branch‐dependent genes were identified using BEAM.

MALDI imaging

The frozen labellum of D. devonianum (25 × 20 × 5 mm) was encased in a 5% carboxymethyl cellulose (CMC) solution and solidified into blocks by freezing at −80 °C for 30 min. The frozen labellum was then sectioned into 80 μm slices at −20 °C. These sections were immediately affixed to glass slides coated with indium tin oxide to facilitate imaging analyses. The sample was sprayed with a matrix solution containing 10 mg/mL 9‐AA in a 1:1 methanol: H₂O (v/v) mixture. Measurements were conducted using an AP‐SMALDI10 high‐resolution MALDI imaging ion source (TransMIT GmbH, Giessen, Germany) coupled with an Orbitrap mass spectrometer (‘Q Exactive’, Thermo Fisher Scientific, Bremen, Germany). The mass spectrometer operated in both positive and negative ion modes, scanning a range of mass‐to‐charge (m/z) values from 50 to 1050 at a mass resolution (R) of 70 000. The measurement speed in full scan mode was approximately 1.5 s per pixel, with a pixel size of 15 μm. Mass spectrometry profiling data were preliminarily viewed and processed using Mirion 3D (v3.3.64.22) software. In negative ion mode, two ion adduct forms, [M − H]⁻ and [M + e]⁻, were considered for metabolite identification, with an acceptable mass error range of ±2 ppm. This approach ensured high‐quality imaging and detailed characterization of the tissue samples (Susniak et al., 2020).

Accession numbers

The raw sequencing data obtained and evaluated in the current study have been deposited in China National GeneBank DataBase (CNGBdb, https://db.cngb.org/) under accession number PRJCA033181. The whole‐genome assembly data have been deposited in CNGBdb under accession number GWHFIDH00000000.1. The genome annotation files, gene CDS, and protein data have been submitted to Figshare (10.6084/m9.figshare.28016453).

Conflict of interest

The authors declare that they have no competing interests.

Author contributions

J.W., Z.X., P.L. and B.D. participated in the conception and design of the research. J.W., Y.Z., M.Z., X.L., H.X., Y.L. and K.W. collected and processed the species. J.W. and T.L. were responsible for analysing, processing data and compiling charts. J.W., Z.X. and B.D. wrote and revised the manuscript. All authors agreed to the submitted version of the manuscript.

Supporting information

Figure S1 The D. devonianum species. (a) The morphological character of floral organs, flower colours and leaf colours in D. devonianum plants. The unique floral organ includes three sepals in the first whorl, three petals in the second whorl and productive parts in the centre of the flower. (c) Morphological anatomy of D. devonianum flowers during floral development. Development stage 1 (S1, bub length 0.2 cm), S2 (bub length 0.4 cm), S3 (bub length 0.6 cm), S4 (bub length 0.9 cm), S5 (bub length 1.0 cm), S6 (bub length 1.1 cm), S7 (bub length 1.3 cm), S8 (bub length 1.5 cm), S9 (bub length 1.7 cm), S10 (bub length 2.0 cm), S11 (bub length 2.3 cm), S12 (slightly blooming flowers), S13 (half blooming flowers), S14 (half blooming flowers), and S15 (full blooming flowers). (c) Morphological features of yellow patches and fringes on the labellum.

Figure S2 Genome size estimation of D. devonianum genome using k‐mer distribution (k = 31).

Figure S3 Hi‐C contact map representing the chromosome‐scale genome of D. devonianum.

Figure S4 Estimation of divergence times for candidate species using MCMCtree. Blue bars represent the 95% Confidence Interval (CI) for divergence times.

Figure S5 KEGG and GO pathway enrichment of significantly expanded (P < 0.01) gene families of the D. devonianum genome.

Figure S6 The analysis of D. devonianum whole‐genome duplications. Synteny analysis between the D. devonianum and other Dendrobium genomes. Dot plots showed the syntenic orthologs between D. devonianum and D. huoshanense (a), between D. devonianum and D. chrysotoxum (b), and between D. devonianum and D. nobile (c). (d) Distribution of K _S for gene pairs recognized by using reciprocal best hits of all‐against‐all BLAST matches. Frequency distributions of K _S values based on paralogous pairs of D. devonianum. (e) Frequency distributions of K _S values based on paralogous pairs of D. nobile. (f) Frequency distributions of K _S values based on paralogous pairs of D. catenatum. (g) Frequency distributions of K _S values based on paralogous pairs of D. chrysotoxum.

Figure S7 Substitution‐rate‐adjusted mixed paralog‐ortholog K _S plot as produced by ksrates. (a) K _S distributions for paralogous pairs and anchored paralogs (blue bars and red bars) identified within Apostasia shenzhenica. The vertical dashed lines labelled ‘a’ and ‘b’ indicate the modes of these components. The mixed paralog‐ortholog K _S plot after each median has been substituted by its original anchor‐pair K _S list. The original anchor‐pair K _S distribution is visible as a grey histogram in the background. (b) Phylogram generated by ksrates from the input phylogenetic tree, with branch lengths set to the K _S distances estimated from ortholog K _S distributions.

Figure S8 Correlation plot with hierarchical clustering of bulk RNA‐Seq and scRNA‐Seq libraries.

Figure S9 (a) UMAP plot showing dimensional reduction of the distribution of flower cells from different stages of development (S1, S6, and S15). Cells with different colours indicated respective clusters. (b) Bar charts show fraction of cells in each cluster of S1, S6 and S15. (c) The number of DEGs in each cell cluster.

Figure S10 Cell‐specific expression profiles of marker genes in the UMAP projection. Expression profiles of well‐known markers identified mesophyll cells, epidermal cell, vascular bundle cells, proliferating cells, xylem, and phloem cells. Colour: Normalized UMI counts. The cell type with enrichment of marker expression has been labelled beside.

Figure S11 (a) Scatter plots of GO enrichment analysis for Cluster 0, 1, 2, 3, 6 and 10. (b) KEGG pathway enrichment analysis of Cluster 1, 3, and 5.

Figure S12 (a, b) UMAP visualization of putative clusters in D. devonianum flower of S1 versus S15 (S1_S15) sample (a) and S1 versus S6 sample (S6_S15) (b). Each dot denotes a single cell. Color denotes corresponding cell clusters. UMAP visualization of putative clusters which shown separately according to S1, S6, and S15. (c) UMAP visualization and cluster distribution of sub‐clusters from S1 versus S6 versus S15 Cluster2 (S1_S6_S15_C2). Each dot indicates a single cell, and the colour of the cells indicates cell identities. (d) Dot plot showing the expression patterns of EC marker genes in S1_S6_S15_C2 sub‐clusters. (e) GO enrichment analysis of the genes in S1_S6_S15_C2 sub‐clusters. Dot size represents ratio of enriched genes; dot colour represents FDR value.

Figure S13 The phylogenetic tree of Type I MADS‐box genes from O. sativa (Os), A. thaliana (AT), P. equestris (Pheq), blueberry (Vc), and D. devonianum (Dede). Type I MADS‐box genes, including 5 Mα and 2 Mγ members exist in the D. devonianum genome. Both P. equestris and D. devonianum lost members in the Mβ clades. Red circles indicate genes in the D. devonianum genome.

Figure S14 The phylogenetic tree of Type II MADS‐box genes from O. sativa (Os), A. thaliana (AT), P. equestris (Pheq), blueberry (Vc), and D. devonianum (Dede). Red circles indicate genes in the D. devonianum genome. Both P. equestris and D. devonianum lost members in the FLC and AGL12 clades.

Figure S15 (a) Heatmap displaying the expression pattern of representative MADS‐box genes in individual floral organs. (b) The flowering ABCE model in D. devonianum that specifies floral organs is proposed based on the gene expression values (bar heights) from (a).

Figure S16 The phylogenetic tree of A, B, C, D, and E‐class genes in orchids.

Figure S17 Expression patterns of MADS‐box genes in scRNA‐seq data. (a) Expression level of MADS‐box genes in different stages of flower development (S1, S6, and S15). (b, c) Expression level of MADS‐box genes in various cell types.

Figure S18 MALDI‐MSI images generated with heatmaps show the accumulation patterns of metabolites in the labellum.

Figure S19 Flavonoids and carotenoid biosynthesis pathway. (a) The flavonoid biosynthetic pathway leading to the biosynthesis of anthocyanins and other flavonoids groups acting as flower pigments. Different coloured circles correspond to different types of flavonoid components. (b) A schematic diagram of carotenoid metabolic pathway. Carotenoid biosynthesis utilizes the plastidial MEP pathway to supply the C5 precursor metabolites IPP and DMAPP.

Figure S20 Expression profile of structural genes involved in the pathway of flavonoid and carotenoid biosynthesis at different areas of the labellum. S15‐la‐Y: Flower15‐labellum‐Yellow, S15‐la‐P: Flower15‐labellum‐Purple, S15‐la‐W: Flower15‐labellum‐White, S15‐la‐F: Flower15‐labellum‐Fringes.

Figure S21 Expression profile of structural genes involved in the pathway of flavonoid and carotenoid biosynthesis in different cell populations at various stages.

Figure S22 (a) KEGG pathway enrichment of differential genes between S1 and S15 periods in EC_f. (b) KEGG pathway enrichment of differential genes between S1 and S15 periods in VB.

Figure S23 Connection networks among main biosynthetic genes. (a) A potential transcriptional regulatory network of TFs with candidate genes related to flavonoids and carotenoid biosynthetics. The larger the circle, the darker the colour, and the greater the degree value. The blue circles represent TFs. The orange circles represent biosynthetic genes. (b) The heatmap of expression of bHLH and MYB. (c) The orange and grey lines represent positive and negative correlations, respectively. The graph size and line width represent the degree of correlation.

Figure S24 Trends in gene expression with branching in cells based on different differentiation fates. Each row represents one gene. These genes were clustered into 5 modules with distinct expression patterns. Different colours represent the gene expression level. The representative genes of each module are shown in the middle panel. The GO terms for each module are shown on the right panel.

Table S1 Statistics of genome sequencing data for genome assembly of D. devonianum.

Table S2 Statistics of the contig‐level and chromosome‐level D. devonianum genome assembly.

Table S3 Assessing the completeness of genome assembly and annotation results for D. devonianum genome with Benchmarking Universal Single‐Copy Orthologs (BUSCO) analysis.

Table S4 Statistics of chromosome‐scale pseudomolecules.

Table S5 Statistics of repetitive element content in the D. devonianum genome.

Table S6 Comparisons of genes and gene families among 16 representative land plants.

Table S7 Statistical data of scRNA‐seq.

Table S8 MADS‐box genes and pigment biosynthetic genes associated with WGD and tandem replication events.

Table S9 Cell number of different clusters identified from the combined sample, shown at S1, S6 and S15 respectively.

Table S10 Locations of MADS‐box genes in D. devonianum.

Table S11 Numbers and categories of MADS‐box genes in published orchid genomes.

Table S12 Locations of flavonoid pathway gene and carotenoid pathway genes in D. devonianum.

Table S13 Statistics of transcription factor (TF) number in D. devonianum genome.

Table S14 The correlation between transcription factors and functional genes.

Table S15 The statistics of Illumina reads mapped to the assembled genome.

Data availability statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.